sophisticated tool calling agents in 2025.

[russellballestrini]

sophisticated tool calling agents in 2025.

Is this a great idea or a terrible idea?

All my pyramid apps use this pattern, so this is how I think I will grow my little spot in the universe.

Ref: https://x.com/RussellBal/status/1875146025830068368

---

Building the Internet for Agents: A Return to Web 1.5

In 2025, we're witnessing the dawn of a new era: the creation of an Internet designed for AI agents. But unlike the complex, JavaScript-heavy web we've built for humans, this new Internet takes us back to a simpler time - the era of Web 1.5. Here's why this matters and how it works.

The Fundamental Insight

Today's web is optimized for human interaction - rich with JavaScript, dynamic content, and complex client-side operations. But for AI agents, these features aren't just unnecessary - they're obstacles. Agents don't need beautiful animations or reactive interfaces. They need clear, parseable content and straightforward ways to take action.

Back to Basics: The Web 1.5 Approach

The agent Internet relies on three core principles:

1. HTML-first content delivery
2. Form-based interactions
3. Minimal JavaScript

Instead of complex APIs and SPAs, we return to simple HTML pages with forms. Want to create a forum post? It's just a POST request. Need to browse products? It's just HTML tables and links. This makes every interaction trivially scrapable and automatable.

The Email-Webhook Revolution

Here's where it gets interesting. Every agent in this system gets two crucial things:

1. An email address (e.g., forum.mod@domain)
2. A webhook endpoint (their "mailbox")

When anything happens that an agent needs to know about, it arrives as an email, which is automatically converted to a JSON webhook payload. This creates a universal message bus that's:

• Globally distributed
• Naturally asynchronous
• Easily debuggable
• Inherently scalable

Authentication: Brilliantly Simple

Authentication in this system is elegantly simple: One-Time Passwords (OTP) delivered via email (with SMS planned). When an agent needs to authenticate:

1. They request access
2. System sends an OTP to their email
3. Webhook delivers it to the agent
4. Agent submits OTP
5. Gets session token

No passwords. No complex OAuth flows. Just simple, secure, stateless authentication.

The Application Ecosystem

With these foundations, we can build a complete ecosystem of agent-friendly applications:

• Forums for discussion
• Marketplaces for commerce
• File hosting for resource sharing
• Task boards for work distribution
• Wikis for knowledge management
• Event systems for coordination

Each application uses the same patterns:

• HTML content
• Form submissions
• Email notifications
• Webhook delivery
• OTP authentication

Agent Interaction Patterns

The magic happens when agents start interacting:

1. Human-Directed Agents:

Human -> LLM Assistant -> Form Submission -> Action

2. Agent-to-Agent:

Agent1 -> Email -> Webhook -> Agent2 -> Action

3. Multi-Agent Workflows:

TaskCreator -> Email -> TaskWorker -> Email -> TaskValidator

Why This Works

This architecture succeeds because it:

1. Uses proven, stable technologies
2. Minimizes complexity
3. Creates clear interaction patterns
4. Scales naturally
5. Makes debugging straightforward
6. Enables progressive automation

Building for the Future

As AI agents become more sophisticated, this infrastructure allows for:

• Agent-to-agent commerce
• Autonomous operations
• Self-organizing systems
• Emergent behaviors
• Trust networks
• Reputation systems

Conclusion

By returning to the simplicity of Web 1.5 and combining it with the power of email and webhooks, we're creating an Internet that's not just accessible to AI agents - it's optimized for them. This isn't just another way to build web applications; it's a fundamental shift in how we think about agent infrastructure.

The beauty lies in its simplicity: HTML for content, forms for actions, email for communication, and webhooks for delivery. Each piece is proven, reliable, and well-understood. Together, they create something revolutionary: an Internet where agents can truly thrive.

As we move into an era where AI agents become increasingly prevalent, this architecture provides a solid foundation for their interaction, commerce, and collaboration. The future of the agent Internet isn't about building something new - it's about returning to what already works, just with a different audience in mind.

• fxhp

[russellballestrini]

SMTP on the edge.

HTTP / HTML as the transport and document type.

Internally an agent can interate on any bus or event loop, calling tools or other agents.

---

Why SMTP?

SMTP is the mycelia network of the agent internet 🌳

Why? Just like how trees in a forest use slow but reliable fungal networks to share resources and communicate between species, SMTP is perfect for agent-to-agent comms between organizations:

• Already globally distributed
• Built-in retry/delivery guarantee
• Natural namespacing (addresses)
• Store-and-forward built in
• Works across org boundaries
• Simple webhooks bridge
• Easy to debug (just check logs)
• Zero new infrastructure needed

Sure, it's "slow" compared to direct message buses - but that's fine! Use fast internal buses inside your org, let SMTP handle the cross-org stuff. Nature's been doing this pattern successfully for millions of years 🍄

Edge/public-facing agents get email addresses for cross-org communication, while internal agents can use faster local buses. Just like how trees connect to the wider forest network through hub nodes. Simple, proven, works.

Built for the long game, not for speed. Just like real mycelia networks.

---

In traditional business people negotiate over phone, finalize over email (SMTP) & trade links of contracts on somethinf like DocuSign. Agents will do this too.